Association of functional connectivity of the executive control network or default mode network with cognitive impairment in older adults with remitted major depressive disorder or mild cognitive impairment

Abstract

Major depressive disorder (MDD) is associated with an increased risk of developing dementia. The present study aimed to better understand this risk by comparing resting state functional connectivity (rsFC) in the executive control network (ECN) and the default mode network (DMN) in older adults with MDD or mild cognitive impairment (MCI). Additionally, we examined the association between rsFC in the ECN or DMN and cognitive impairment transdiagnostically. We assessed rsFC alterations in ECN and DMN in 383 participants from five groups at-risk for dementia—remitted MDD with normal cognition (MDD-NC), non-amnestic mild cognitive impairment (naMCI), remitted MDD + naMCI, amnestic MCI (aMCI), and remitted MDD + aMCI—and from healthy controls (HC) or individuals with Alzheimer’s dementia (AD). Subject-specific whole-brain functional connectivity maps were generated for each network and group differences in rsFC were calculated. We hypothesized that alteration of rsFC in the ECN and DMN would be progressively larger among our seven groups, ranked from low to high according to their risk for dementia as HC, MDD-NC, naMCI, MDD + naMCI, aMCI, MDD + aMCI, and AD. We also regressed scores of six cognitive domains (executive functioning, processing speed, language, visuospatial memory, verbal memory, and working memory) on the ECN and DMN connectivity maps. We found a significant alteration in the rsFC of the ECN, with post hoc testing showing differences between the AD group and the HC, MDD-NC, or naMCI groups, but no significant alterations in rsFC of the DMN. Alterations in rsFC of the ECN and DMN were significantly associated with several cognitive domain scores transdiagnostically. Our findings suggest that a diagnosis of remitted MDD may not confer functional brain risk for dementia. However, given the association of rs-FC with cognitive performance (i.e., transdiagnostically), rs-FC may help in stratifying this risk among people with MDD and varying degrees of cognitive impairment.

Introduction

A large body of literature supports that major depressive disorder (MDD) is associated with an increased risk of developing dementia in general and Alzheimer’s dementia (AD) in particular [1]. Current or remitted MDD in older adults is frequently accompanied by mild cognitive impairment (MCI), and this comorbidity is thought to increase the risk for future dementia [2, 3]. A proposed mechanistic model linking MDD and dementia suggested that patients with MDD present with abnormalities in (1) the frontal-executive circuit subserving the executive control network (ECN), and (2) the corticolimbic circuit subserving the default mode network (DMN); in turn, these abnormalities could diminish cognitive reserve and increase the risk for dementia or AD [4].

We previously studied the structural brain alterations of white matter tracts and gray matter cortical thickness in frontal-executive and corticolimbic circuits in five non-overlapping groups at-risk for AD: (1) remitted MDD with normal cognition (MDD-NC); (2) non-amnestic MCI (naMCI) with no history of MDD; (3) MDD + naMCI; (4) amnestic MCI (aMCI) with no history of MDD; and (5) MDD + aMCI. Our study also included two control groups: patients with AD and non-psychiatric (“healthy”) controls (HC) [5]. Our results showed that older individuals with remitted MDD or remitted MDD + MCI did not have the same white or gray matter changes in the frontal-executive and corticolimbic circuitries as those with aMCI or AD, suggesting distinct neural risk mechanisms in these diagnostic groups. Building on these results, we aimed to examine alterations of resting state functional connectivity (rsFC) of ECN and DMN in the same seven diagnostic groups.

Previous resting state functional MRI (rs-fMRI) studies that have shown alterations in the within- or between-network functional connectivity of ECN [6, 7], DMN [8, 9], or both [10, 11], or absence of differences [12, 13], in older adults with MDD vs. healthy older adults. Only a few studies have included additional diagnostic groups such as MCI and MDD + MCI [14–17]. However, these studies had small sample sizes and they often included heterogeneous samples of MDD (e.g., samples including patients with remitted, acute, and treatment-resistant episodes) and MCI (e.g., samples including patients with amnestic and non-amnestic MCI). The literature is thus inconclusive [18].

To address some of these limitations, the current study focused on our five at-risk groups with larger Ns plus two control groups, participants with AD and HCs. Since aMCI (compared with naMCI) is more likely to progress to AD [19], and having both MDD and MCI is thought to confer a higher risk for dementia or AD than either disorder individually [20–22], we expected that neuroimaging biomarkers indicating a risk for dementia should differ in our groups according to the hypothesized risk for dementia. Thus, we hypothesized that alteration of rsFC (hypo or hyperconnectivity) in the ECN and DMN would be progressively greater among our seven groups, ranked as HC, MDD-NC, naMCI, MDD + naMCI, aMCI, MDD + aMCI, and AD.

Additionally, using a transdiagnostic approach in the five at-risk groups and HC, we explored the association between alterations of rsFC in the ECN and DMN and performance in specific cognitive domains (i.e., executive functioning, processing speed, language, visuospatial memory, verbal memory, and working memory). Previous rs-fMRI studies have not assessed alterations of brain networks associated with specific cognitive deficits in individuals at-risk for AD [23]. In doing so, we aimed to explore the common and distinct neurobiological mechanisms that underpin cognitive dysfunction across the population at-risk for AD. We propose that cognitive performance related to executive function and memory performance respectively would be dimensionally associated with altered within- or between-network rsFC of ECN and DMN.

Methods

Participants

Overall, 414 participants across the seven diagnostic groups were considered; after quality check (QC; see “MRI data preprocessing and quality check (QC)” below) of their scan, 383 were included in the analyses. Participants in the five groups at-risk for dementia were recruited in the Prevention of Alzheimer’s Dementia with Cognitive Remediation Plus tDCS in Mild Cognitive Impairment and Depression trial (PACt-MD; ClinicalTrials.gov Identifier: NCT02386670), approved by the institutional review board of the Centre for Addiction and Mental Health (CAMH), in Toronto, Canada. All PACt-MD participants provided written informed consent and underwent detailed clinical and neuropsychological (NP) assessments, as described previously [24, 25]. Of the PACt-MD participants, 291 underwent a rs-fMRI scan; 3 were excluded because they received a consensus diagnosis of dementia [25]; 13 more were excluded after QC of the rs-fMRI scans; and the remaining 275 are included in this analysis. Their baseline diagnoses were adjudicated according to the criteria of the Diagnosis and Statistical Manual 5th edition (DSM-5), as well as the National Institute of Aging-Alzheimer Association 2011 criteria [26] during a consensus conference, reported previously [24, 25]. In brief, of the 275 participants in the five non-overlapping groups—MDD-NC (n = 49); naMCI (n = 42); aMCI (n = 101); MDD + naMCI (n = 32); MDD + aMCI (n = 51)—those with remitted MDD (with or without MCI) were over the age of 65 and had a history of one or more major depressive episode (MDE), in remission for at least 2 months. Participants with aMCI and naMCI (without MDD) were older adults over the age of 60, with no history of MDE their adult life. For MCI subtypes, if the participants had impaired memory on the NP battery, they were defined as aMCI. If they did not, they were classified as naMCI. For further details of the inclusion and exclusion criteria see Supplementary Table 1.

Additionally, the 56 patients with AD, were 55 years and older, had participated in two clinical trials of AD [27, 28], and had completed a 3T rs-fMRI scan. Ten were excluded after QC of their scan, leaving 46 in the AD group. Finally, of 80 HC participants with normal cognition and no psychiatric history who had taken part in one of four studies (i.e., PACt-MD (n = 34), the two AD trials (n = 22); or a cohort study of normal aging (n = 14)) and had a baseline 3T rs-fMRI scan, eight were excluded after QC of scans, leaving 62 in the HC group.

MRI acquisition and analysis

All participants were scanned using a 3T GE Echospeed (General Electric Medical Systems, USA) research-dedicated scanner at CAMH, with an 8-channel head coil using the same multimodal MRI protocol. During the scanning, participants were instructed to lie still with their eyes open. Functional images were obtained axially using a Spiral-in/out acquisition method [29]. The following parameters were applied for a 7-min functional imaging run: TR (repetition time) = 2000 ms, TE (echo time) = 30 ms, field of view (FOV) = 20 cm, Flip angle = 77◦, Slice thickness = 4 mm, 40 slices.

T1-weighted scans were acquired as a sagittal 3D fast spoiled gradient echo image with 0.9 mm isotropic voxels (TE: 3 ms; TR: 8.2 ms, flip angle 8, FOV = 24 cm). Reconstruction of the cortical surfaces was performed using the FreeSurfer toolkit version 6.0.0 [30, 31].

MRI data preprocessing and quality check (QC)

First, an in-house pipeline, FeeINCS, was used to automatically evaluate and remove artifacts induced by the spiral in/out acquisition (https://github.com/TIGRLab/FeenICS). FeeINCS includes realignment for motion via FSL’s MELODIC; parameters were used to calculate framewise displacement (FD) as the sum of the absolute values of the six head motion parameters of (x-, y-, z-) translations and (pitch-, yaw-, roll-) rotations [32, 33]. Scans with a mean FD greater than 0.5 mm and/or if over 25% of FDs were above 0.5 mm, equal to less than 5 min good data [33] were excluded from subsequent analyses (leaving n = 383 scans—see above). Data were then formatted into the Brain Imaging Data Structure (Gorgolewski et al. [34]; http://hub.docker.com), and preprocessing (Slice time correction, Pre-processed BOLD in native space, EPI to T1w registration, Resampling BOLD runs onto standard spaces, EPI sampled to FreeSurfer surfaces) was performed with fMRIPrep 1.3.2 [35], based on Nipype 1.1.9 [34, 36].

Ciftify and surface-based data

To adjust for the inherent geometry of individual cortical surfaces and to boost the statistical power, the ciftify pipeline was used to transform the images from the FreeSurfer format to the surfaces based Connectivity Informatics Technology Initiative (CIFTI) format ([37]; https://github.com/edickie/ciftify). First, the individual T1-weighted data were transferred to CIFTI format and then, the inter-subject anatomy-based were registered to the MNI space using the ciftify_recon_all function. Next, the rs-fMRI data were projected to the surface and subcortical data were resampled, using cifitfy_subject_fmri function. Next, ciftify_clean_img was applied for temporal filtering (0.01–0.08 Hz), nuisance regression (i.e., six head motion parameters, mean global signal, cerebrospinal fluid (CSF) and white matter signals), and surface based smoothing data with Gaussian kernel of 6 mm full width at half maximum.

Functional connectivity analyses: dual regression and resting state networks (RSNs)

The dual-regression technique uses the group-ICA (independent component analysis) spatial maps for back-reconstruction purposes. Following previous studies [38, 39], we used the predefined templates of ECN and DMN from the ten resting-fMRI components identified in a prior study [40, 41]; https://www.fmrib.ox.ac.uk/datasets/brainmap+rsns/). The ECN and DMN templates were projected to the surface and subcortical masks of each participants’ T1-weighted image using ciftify_vol_result function, and all the CIFTI files were averaged together to create a surface-based common template for each network. These ECN and DMN surface-based common templates were used as the group-ICA maps in the dual regression analysis. Each common template was mapped to each participant’s processed rs-fMRI data (4D dataset), and a mean time series was extracted as the average of the timeseries of all vertices within each network (i.e., ECN, DMN). Next, the mean time-series were regressed into the same 4D dataset, by modeling the linear association between the mean time-series of each network (i.e., ECN, DMN) and the time series of each single vertex, resulting in a set of subject-specific whole-brain functional connectivity maps for each network [42, 43].

Statistical analysis

Whole brain network connectivity maps were used to examine differences across groups. To test our hypothesis, contrasts were set to a linear relationship: HC > MDD-NC > naMCI > MDD + naMCI > aMCI > MDD + aMCI > AD, and HC < MDD-NC < naMCI < MDD + naMCI < aMCI < MDD + aMCI < AD. Permutation Analysis of Linear Models (PALM) with 5000 permutation testing was used, controlling for age, sex, education, and mean FD [44]. To adjust for multiple comparisons, we used threshold-free cluster enhancement across cortical vertices (TFCE) [45] and group results were set at threshold of p < 0.05 FDR-corrected for the number of vertices in each hemisphere, and we further corrected for separate runs of PALM for each hemisphere (critical level a = 0.025). If the main effect of diagnostic group was significant, a binary mask was used to extract the significant parameter estimates from the subject-specific functional connectivity maps. Next, post hoc Tukey test and Cohen’s d values were calculated using R. Differences within the network templates were considered as within-network connectivity, while differences outside the template were labeled as between-network (i.e., representing connectivity between the DMN or ECN network and other brain networks).

Association between brain networks rsFC and cognitive domain performances

Among the PACt-MD participants, 271 of 275 in the at-risk groups and all 30 HC completed the comprehensive NP battery [24]. The characteristics of the composite scores across the five at-risk groups and the HC group from the PACT-MD study (n = 301) are presented in Supplementary Table 2. For each of these 301 participants, domain scores were created for six cognitive domains (executive functioning, processing speed, visuospatial memory, verbal memory, working memory, and language) (Fig. 1). Z-scores were calculated for each cognitive measure using the mean and standard deviation (SD) of all the PACt-MD HC participants (n = 81; with or without collected rs-fMRI scan) and domain scores were created by grouping measures into domains and averaging the Z-scores into a single domain score. Extreme scores were capped (winsorized) at ±5 SD.

Fig. 1. Cognitive domain scores and their components.

The composite scores for each domain were created from averaging Z-scores of multiple related tests (see Supplementary Table 3). TMT B/A (Trails-Making Test part B; central executive processes/A; complex attention), CDT (clock-drawing test), SCWT (Stroop Color and Word Test), Coding (Digit-Symbol-Coding from Wechsler Adult Intelligence Scale-III), BNT (Boston Naming Test—split form), Sem. Flu. (semantic fluency), Letter Flu. (letter fluency), BVMT-R (Brief Visuospatial Memory Test–Revised), CVLT-II (California Verbal Learning Test-II ), d‘ (d prime), % ret. (percent retained), CPT-IP (Continuous Performance Test-Identical Pairs version), PASAT (Paced Auditory Serial Addition Task).

To assess the association between brain connectivity and cognition, we used a transdiagnostic approach (n = 301). Each of the domain scores was regressed on the ECN and DMN functional connectivity maps as a continuous variable, controlled for age, sex, education and mean FD. We used FSL’s PALM package, using 5000 permutations with the TFCE approach [45]. Group results were set at an FDR-corrected threshold at p < 0.05 (corrected for the number of vertices in each hemisphere), and critical level was set to 0.025 to further correct for separate runs of PALM for each hemisphere. If the primary regression model was significant, a post hoc test for main effect of diagnostic group was conducted in the extracted parameter estimates from the subject-specific functional connectivity maps.

Results

Demographics

Demographics and clinical characteristics of the 383 participants are presented in Table 1. The mean age in the AD and aMCI groups was significantly older than in the HC group, and the mean level of education in the AD group was lower than in all other groups. Most participants with MDD (with or without MCI) were taking a serotonergic antidepressant at the time of study.

Table 1.

Demographics and clinical characteristics.

rs-fMRI study participants (N total = 383)	HC (N = 62)	MDD (N = 49)	naMCI (N = 42)	MDD + naMCI (N = 32)	aMCI (N = 101)	MDD + aMCI (N = 51)	AD (N = 46)	Statistics, p value	Post hoc results
Age Mean ± SD	70.1 ± 6.1	70.6 ± 4.8	71.6 ± 6.6	71.1 ± 4.4	73.5 ± 7.8	71.6 ± 4.5	75.2 ± 7.0	F(6,344) = 3.4 p = 0.002	AD > HC, MDD aMCI > HC
N (%) Female	43/19 (69.3%)	33/16 (67.3%)	22/20 (52.3%)	22/10 (68.7%)	54/47 (53.4%)	29/22 (56.8%)	28/18 (60.8%)	χ2(6) = 7.4, p = 0.2	–
Education Mean ± SD	6.3 ± 0.9	6 ± 1.1	5.9 ± 1.0	6.2 ± 0.7	5.7 ± 1.2	5.5 ± 1.1	5.2 ± 2.0	F(6,376) = 4.3, p = 0.0002	AD < HC, MDD + naMCI MDD + aMCI < HC
Race
Caucasian (%)/	83.3%	91.6%	65.8%	83.8%	73.2%	78%	58.6%	–	–
Black (%)/	0%	2.1%	7.3%	3.2%	6.9%	8%	6.5%
Asian (%)/	12.5%	0%	14.6%	3.2%	13.8%	8%	13.0%
Other (%)	4.2%	6.2%	12.1%	9.6%	5.9%	6%	21.7%
MMSE scoresa Mean ± SD	29.1 ± 1.0	29.0 ± 1.1	28.4 ± 1.3	28.6 ± 1.3	27.4 ± 1.7	27.6 ± 1.8	23.1 ± 3.1	F(6,371) = 63.5,      p < 2e−16	AD < all groups. aMCI < HC, MDD, MDD + naMCI, naMCI MDD + aMCI < HC, MDD
Mean FDb Mean ± SD	0.12 ± 0.06	0.13 ± 0.05	0.12 ± 0.06	0.12 ± 0.05	0.14 ± 0.07	0.15 ± 0.07	0.16 ± 0.07	F(6,376) = 2.3, p = 0.03	Not sig
Age at first onset of MDD (years) Mean ± SD	NA	39.0 ± 18.3 (n = 49)	NA	36.5 ± 17.9 (n = 32)	NA	42.2 ± 17.4 (n = 51)	NAh	F(2,126) = 1.0, p = 0.3	–
N (%) recurrent MDDc	NA	40 (81.6%)	NA	25 (80%)	NA	33 (67.3%)	NA	X2 = 3.2, df = 2, p = 0.1	–
Number of MDEsd Mean ± SD	NA	4.0 ± 1.4	NA	4.2 ± 1.7	NA	3.8 ± 1.7	NA	F(2,126) = 1.0, p = 0.3	–
Current use:	NA		NA		NA		NA	–	N = 29 missing values
SSRIs		28		13		30
Non-SSRIs		14		8		22
Previous use:
SSRIs		33		16		35
Non-SSRIs (n’s)e		21		10		30
EOD/LOD (n’s)f	NA	30/8	NA	24/3	NA	36/6	NA	–	N = 25 missing values
Time since remission of last major depressive episode (years) (n = 116) Mean ± SD	NA	9.3 ± 12.8	NA	5.4 ± 7.3	NA	10.4 ± 6.7	NA	–	N = 16 missing values
MADRSg Mean ± SD	1.1 ± 1.6	4.9 ± 3.7	3.5 ± 2.2	5.0 ± 3.0	3.3 ± 2.5	4.4 ± 3.2	NA	F(5, 249) = 6.7, p = 6.0e−06	MDD > HC, aMCI, naMCI MDD + naMCI > HC, naMCI MDD + aMCI > HC, naMCI

^aMini-mental state examination.

^bFramewise displacement.

^cRecurrent major depressive episodes (MDE) vs. single MDE.

^dMajor depressive episode (MDE).

^eSerotonergic-noradrenergic reuptake inhibitor, tricyclic antidepressants, atypical antidepressants (e.g., bupropion), aripiprazole.

^fEarly-onset depression (EOD; onset of depression prior the age of 60)/late-onset depression (LOD; onset after the age of 60).

^gMontgomery-Asberg Depression Rating Scale.

^hData not available.

ECN and DMN resting state functional connectivity (rsFC) across the seven groups

No significant effects of the linear group contrast were found in the DMN. There was no significant difference in within-network rsFC of the ECN across diagnostic groups. However, we found significant effects between ECN and two other networks: the pre and post central gyrus (somatosensory network (SMN)) and parts of supramarginal gyrus (SMG) [ECN-SMN-SMG], across the seven groups (Fig. 2a). After extracting connectivity values for the observed region, the main effect of diagnostic group was F(6, 372) = 3.7, corrected p < 0.025, η² = 0.06. In the post hoc Tukey test, the AD group showed lower anti-correlation between ECN-SMN-SMG compared to the HC, MDD-NC, and naMCI groups (p < 0.05). No differences were found involving the MDD + naMCI, aMCI, and MDD + aMCI groups (Fig. 2b).

Fig. 2. Alterations of the executive-control network resting state functional connectivity with other brain networks.

a Significant alterations in the functional connectivity between ECN and the somatosensory network (SMN) and parts of supramarginal gyrus (SMG) (ECN-SMN-SMG). Displayed results are FDR corrected significant (p < 0.05), with Bonferroni correction for two hemispheres (critical level a = 0.025). b The AD group showed decreased anti-correlation compared with the HC, MDD and naMCI groups (p < 0.05).

Association between ECN and DMN rsFC and cognition

Processing speed was negatively correlated with the rsFC between ECN and bilateral SMN (cluster FDR corrected p = 0.003, R = −0.3). That is, a higher anti-correlation between ECN-SMN was associated with better processing speed (Fig. 3a).

Fig. 3. Association of executive-control network resting state functional connectivity with processing speed and language.

a Significant association between processing speed and the rsFC between ECN and bilateral sensorimotor network (SMN). b Significant association between language and the rsFC between ECN and right inferior parietal lobule (IPL) part of the auditory network (AN), and right transverse temporal (part of the right SMN). Displayed results are FDR corrected significant (p < 0.05), with Bonferroni correction for two hemispheres (critical level a = 0.025).

Language performance was positively correlated with the rsFC between ECN and right inferior parietal lobule part of the auditory network (AN), and right transverse temporal (part of the right SMN) rsFC (cluster FDR corrected p = 0.01, R = 0.27) (Fig. 3b). That is, a higher correlation between ECN-AN-SMN is associated with better performance in language. The results for the post hoc Tukey test for the diagnostic groups is provided in the Supplementary Material.

Working memory was positively correlated with the rsFC between DMN and right insular (Ins) and superior temporal gyrus (STG) (part of AN) (cluster FDR corrected p = 0.004, R = 0.28). That is, greater correlation between DMN-Ins-STG is associated with better working memory performance (Fig. 4a).

Fig. 4. Association of default mode network resting state functional connectivity with working memory and visuospatial memory.

a Significant association between working memory and the rsFC between DMN and right insular (Ins) and superior temporal gyrus (STG). b Significant association between visuospatial memory and the rsFC between DMN and left supramarginal (SMG) and superior parietal lobe (SPL) part of the left frontoparietal network (L-FPN). Displayed results are FDR corrected significant (p < 0.05), with Bonferroni correction for two hemispheres (critical level a = 0.025).

Visuospatial memory was negatively correlated with the rsFC between DMN and left supramarginal (SMG) and superior parietal lobe (SPL) part of the left frontoparietal network (cluster FDR corrected p = 0.01, R = −0.29). That is, greater anti-correlation between DMN-SMG-FPN is associated with better visuospatial memory performance (Fig. 4b). The results for the post hoc Tukey test within each diagnostic group is provided in the Supplementary Material.

We found no association between executive functioning and verbal memory composite scores and ECN or DMN rsFC.

Discussion

We used dual regression analysis to assess rsFC alterations in the ECN and DMN in five groups of participants at-risk for AD (MDD-NC, naMCI, MDD + naMCI, aMCI, MDD + aMCI), and two additional groups, HC and patients with mild-to-moderate AD. We found significant alteration of rsFC between ECN and the SMN and parts of the SMG across the seven groups. As expected [46, 47], this finding was driven by differences between the AD group and the HC, MDD-NC, and naMCI groups, suggesting impairment of the between-network functional integrity in the AD group. Integrations of cognitive related systems (ECN) and sensory motor system (SMN) are essential for perceiving and responding to information, and dysfunction in between-network functional integration will lead to impaired function at the clinical level [48]. In the DMN, we found no significant alterations of within- or between-network rsFC within the seven groups. In addition, we used a trans-diagnostic approach in the five at-risk groups, and HC participants, and found alterations of rsFC of ECN and DMN were associated with several cognitive domain scores consistent with current conceptualizations of brain network function.

The lack of significant effect of diagnostic groups in the within-network rsFC of ECN or DMN or between-network rsFC of DMN was contrary to our hypothesis and some previous studies [8, 10, 15, 17, 49–51], but similar to some other studies that have reported the absence of differences between: HC vs. MDD [12, 13], HC vs. naMCI [19, 52], HC vs. aMCI [53, 54], aMCI vs. AD [55, 56], or AD vs. HC [57]. Only a few studies with more than two comparison groups have reported significant findings, using a variety of analysis methods, different from our methods [15–17, 53, 58]. Further, previous studies in DMN rsFC have reported both hyperconnectivity[59, 60] and hypoconnectivity [61, 62] in AD patients compared to the HC. These changes may reflect direct or indirect pathologic mechanisms and functional disconnection and compensation in response to the damage at earlier stages of neurodegeneration [63]. Additionally, while there is a strong tendency in the literature toward examining the DMN, the lack of negative results, could be suggestive of publication bias [64].

There are several factors that could explain the lack of difference across groups, especially among those with remitted MDD. First, exposure to serotonergic antidepressants can alter rsFC of ECN or DMN [65–67], and more than three quarters of our remitted MDD participants were taking a serotonergic antidepressants. Second, remission of depressive symptoms in older adults with MDD leads to partial restoration of rsFC in the DMN [68, 69], and all our MDD participants were in remission. Further, MCI in remitted MDD might differ from MCI in the absence of MDD [70]. Third, distinct neural mechanisms have been reported in early-onset depression (EOD; onset of depression prior to age 60) vs. late-onset depression (LOD; onset over age of 60) [71–75]. For instance, CSF hallmarks of AD have been found in LOD [76], and nearly all our MDD participants had EOD. Moreover, the pattern of cognitive impairment may be different in EOD and LOD [77]. Fifth, it has been shown that proximity of depressive symptoms to cognitive impairment, especially within 2 years, is associated with increased risk of AD [78]. However, the mean time since offset of last MDE was over 5 years in our MDD group. Lastly, we cannot rule out a type II error, and acknowledge heterogeneity within our diagnostic groups. Taken together, our findings suggest that older adults with early-onset remitted MDD, whose symptoms have been well-managed for some time, may not be as much at-risk for AD as previously thought. Future studies with larger populations could clarify whether subsets of participants with the same diagnoses show different patterns of functional brain change.

In the secondary analysis, while the lack of association between executive functioning and verbal memory composite scores with ECN or DMN rsFC, were contrary to previous findings [9, 79], taken together, all four of our rs-FC neurocognitive associations are anatomically consistent with the conceptualizations of brain circuit functions in relation to cognitive performance (i.e., processing speed and language with ECN, working memory and visuospatial memory with DMN). We found that: (1) faster processing speed was associated with higher rsFC anti-correlation between the ECN and bilateral SMN. This could be interpreted as greater network segregation between the ECN and SMN supporting faster processing speed. Previous findings in heathy older adults have shown that faster processing speed could predict a higher within-network FC of SMN, and lower between-network FC between SMN and other networks including ECN [80]. Impaired processing speed has been identified as an early predictor of progression to AD [81–83], but is also a common finding in aging in general [84]. (2) Better language performance was associated with higher correlation between ECN-right AN-right SMN. This is in line with previous findings in healthy aging, that suggested a compensatory mechanism between ECN-AN [85, 86] and involvement of the right SMN [87] for maintaining a sufficient level of language performance. (3) Better performance in working memory was associated with higher correlation between DMN-right insular-superior temporal (part of AN). Previous working memory task-fMRI studies have shown increased deactivation of posterior DMN in MCI [88], and decreased deactivation of DMN in remitted MDD [89]. However, in healthy younger adults, a better performance in working memory is associated with higher anti-correlation between DMN-AN [90–92]. (4) Better performance in visuospatial memory was associated with higher anti-correlation between DMN-left supramarginal (SMG)-SPL, consistent with task fMRI studies [93, 94]. However, post-hoc analyses showed that while the same associations were present within most diagnostic groups, they were not present within all groups. Nevertheless, the directionality of the association was the same within all six groups, collectively driving robust associations across all. Some participants may use compensatory networks or other networks to perform the same cognitive tasks. Using task-based functional neuroimaging, we and others have shown that there is variability among individuals via engagement of different networks during the same cognitive tasks [95, 96].

Overall, our results show that alterations associated with AD in the ECN between-network rsFC are not present in those with remitted MDD. This suggests rsFC is not enough to identify older adults with remitted MDD who are at risk of developing AD. It is also possible that antidepressant medications might delay or rescue functional brain impairments. Alterations of between-network rsFC of ECN or DMN, associated with presence of cognitive impairments, may have clinical implications in the groups at-risk for AD. Recent studies suggest that between-network rsFC might serve as a reliable biomarker for early detection of AD [97, 98]. Increased alterations of between-network rsFC (vs. within-network) might be due to reduced specificity of neuronal responses [99, 100], or a compensatory mechanism [101, 102]. Using an approach mapping brain circuit function onto cognitive impairment, independent of diagnosis, might serve ultimately to better identify individuals at risk for AD.

Strengths and limitations

Our study has unique strengths; first, we included seven groups across a spectrum from HC to AD. Second, data from comprehensive clinical and cognitive assessments were used during a consensus conference to classify at-risk participants in five diagnostic groups based on diagnostic criteria. Third, we used cognitive domain scores, which reduces measurement error and are more reliable than single test scores [103]. Finally, we examined brain network FC and their association with cognitive performance to provide a comprehensive picture of functional brain alterations across the at-risk groups. With respect to limitations, we acknowledge the possible effects of antidepressant medications, and all participants with MDD having remitted MDD and mostly EOD. Further, all PACt-MD participants were highly motivated to enroll in a clinical trial and may not represent the typical patients with remitted MDD or MCI. Finally, the lack of comprehensive cognitive assessments some of the HC or AD participants could have decreased our power in detecting all the associations between brain networks rsFC and cognitive impairment related to AD.

Conclusion and future directions

Overall, we found significant alterations of rsFC between ECN-SMN, but no significant changes in the rsFC of DMN across the seven groups. Lack of significant findings could be due to the heterogeneity of the MDD and MCI groups (population at-risk of AD), or a type II error. Although speculative, it is also possible that successful treatment of MDD in older adults may prevent or restore brain functional alterations associated with AD. Further, our findings suggest that alterations of between-network rsFC of ECN, and DMN in the presence of cognitive impairment may be an early neuroimaging diagnostic biomarker in the groups at-risk of AD.

Future research should use larger samples and focus on investigating the longitudinal effect of depression (remitted vs. acute MDD, treatment-resistant MDD, or EOD vs. LOD), and risk for different types of dementia including vascular, Lewy body, and AD using multi-imaging modalities.

Author contributions

NRR: substantial contributions to the conception or design of the work and the acquisition, analysis, or interpretation of data for the work; drafting the work or revising it critically for important intellectual content; final approval of the version to be published; and agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. TKR, CH and BGP: substantial contributions to the conception or design of the work; the acquisition, analysis, or interpretation of data for the work; and final approval of the version to be published. SK and NH: substantial contributions to the conception or design of the work; or the acquisition, analysis, or interpretation of data for the work; and final approval of the version to be published. LM, AJF, CEF, MAB, EWD, CRB and MS: substantial contributions to the acquisition, analysis, or interpretation of data for the work; and final approval of the version to be published. BHM and ANV: substantial contributions to the conception or design of the work; the acquisition, analysis, or interpretation of data for the work; drafting the work or revising it critically for important intellectual content; final approval of the version to be published; and agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Funding

This Project has been made possible by Brain Canada through the Canada Brain Research Fund, with the financial support of Health Canada and the Chagnon Family.

Competing interests

NRR has received funding by the Alzheimer Society of Canada Research Program (ASRP) Doctoral Fellowship. ANV has received grant support from the Canadian Institutes of Health Research (CIHR), National Institute of Mental Health (NIMH), the Brain and Behavior Research Foundation (BBRF), Canada Foundation for Innovation (CFI), the CAMH Foundation, and University of Toronto. AJF has received grant support from the US National Institutes of Health, the Patient-Centered Outcomes Research Institute, the Canadian Institutes of Health Research, Brain Canada, the Ontario Brain Institute, and the Alzheimer’s Association. BHM holds and receives support from the Labatt Family Chair in Biology of Depression in Late-Life Adults at the University of Toronto. He currently receives research support from Brain Canada, the Canadian Institutes of Health Research, the CAMH Foundation, the Patient-Centered Outcomes Research Institute (PCORI), the US National Institute of Health (NIH), Capital Solution Design LLC (software used in a study founded by CAMH Foundation), and HAPPYneuron (software used in a study founded by Brain Canada). Within the past 5 years, he has also received research support from Eli Lilly (medications for a NIH-funded clinical trial) and Pfizer (medications for a NIH-funded clinical trial). He has been an unpaid consultant to Myriad Neuroscience. BGP receives research support from the Peter & Shelagh Godsoe Endowed Chair in Late-Life Mental Health, CAMH Foundation and Discovery Fund, National Institute of Aging, Brain Canada, the Canadian Institutes of Health Research, the Alzheimer’s Drug Discovery Foundation, the Ontario Brain Institute, the Centre for Aging and Brain Health Innovation, the Bright Focus Foundation, the Alzheimer’s Society of Canada, the W. Garfield Weston Foundation, the Weston Brain Institute, the Canadian Consortium on Neurodegeneration in Aging and Genome Canada. BGP receives honoraria from the American Geriatrics Society and holds United States Provisional Patent Nos. 16/490,680, 17/396,030 and Canadian Provisional Patent No. 3,054,093 for a cell-based assay and kits for assessing serum anticholinergic activity. CEF receives grant funding from the following sources: CCNA/CIHR, Vielight Inc, Hoffman La Roche, NIH, Brian Canada, St. Michaels Hospital Foundation Heather and Eric Donnelly Endowment. EWD has received funding from BBRF, NIMH, CHIR and CAMH Foundation. CH has received funding from Canadian Institutes of Health Research (CIHR), National Institute of Mental Health (NIMH), and the CAMH foundation. SK has received grant support from Brain Canada, NIH, BBRF, BrightFocus Foundation, Weston Brain Institute, Canadian Centre for Ageing and Brain Health Innovation, CAMH foundation, and University of Toronto, and in-kind equipment support from Soterix Medical Inc. TKR has received research support from Brain Canada, Brain and Behavior Research Foundation, BrightFocus Foundation, Canada Foundation for Innovation, Canada Research Chair, Canadian Institutes of Health Research, Centre for Aging and Brain Health Innovation, National Institutes of Health, Ontario Ministry of Health and Long-Term Care, Ontario Ministry of Research and Innovation, and the Weston Brain Institute. TKR also received in-kind equipment support for an investigator-initiated study from Magstim, and in-kind research accounts from Scientific Brain Training Pro, and participated in 2021 in one advisory board meeting for Biogen Canada Inc. TKR is also an inventor on the United States Provisional Patent No. 17/396,030 that describes cell-based assays and kits for assessing serum cholinergic receptor activity. CRB receives in-kind research support from Scientific Brain Training Pro. He has had grant support from Lundbeck, Pfizer, and Takeda. LM, NH, MAB, and MS have no potential conflicts of interest to declare.

Supplementary information

The online version contains supplementary material available at 10.1038/s41386-022-01308-2.